The Telco Benchmark
Telecom billing and rating application benchmark has been used as a case study to demonstrate the performance gains of incorporating DFPA hardware acceleration solutions. This application is typically a module of BSS/OSS software. The three fold merit of hardware DFPA acceleration can be explained as follows.
In terms of speed: Prepaid services, real time charging (per second charge is much more computation intensive), and the need for convergent billing (common management methodology of all users and all services); all induce tough computational issues. Quoting HP’s Nigel Upton, General Manager for BSS, Communication, and Media: “Real-time charging is the key that unlocks the revenue potential of a new generation service.” In turn, speed is a key to enable feasible real time charging at very high call rates and fine charge resolutions.
In terms of accuracy: Telecommunications services providers and other utility companies suffer from revenue losses of 2%~4% due to the rounding errors induced by binary computations [Source: Telarix]. DFPA accelerators fully alleviates the binary floating point inaccuracy issues and consequently eliminates related revenue losses.
In terms of energy: The computation intensive nature of BSS/OSS data centers means that they are heavy energy users, which results also in an excess demand for cooling. The leading mobile communications services provider Vodafone has made a public statement of responsibility regarding the positive impact of deploying green technology on the environment: “Our Energy Efficiency and Green Technology Steering Team is responsible for ensuring our strategy on energy and climate change is implemented across our technology operations.” [Source: Vodafone]
Benchmark Results
The test was implemented using an AMD Athlon II X2 processor, at 2.8 GHz, with 2 GB of RAM–running Red Hat Enterprise Linux (RHEL) 5.3 64-bit operating system. It was driven by IBM’s “Telco Benchmark”. The benchmark provides an input data stream representing the call durations in seconds of a total of 1 million telephone calls to drive the test. For each call, a charging rate is selected (e.g. local or long distance), call toll is calculated, and taxes are added; then results are rounded to two decimal fraction digits.
The benchmark has resulted in an overall processing performance boost of about 600% (including Disk I/O and number-to-string conversion times). The net computation time has been slashed by a factor of about 75 times. While the computation time initially dominated the overall benchmark processing time by an 85% factor (without decimal acceleration), it has introduced only a fraction of 6% of the later when the SilMinds’ accelerator card was plugged in.